☝️Abstract

Digital currencies are widely recognized as financial assets and mediums of exchange. Cryptocurrency transactions have attracted the attention of investors due to the potential for high returns on cryptocurrency investments. To optimize the profitability of cryptocurrency investments, accurate price predictions are crucial. In recent years, various modern technologies and statistical models have been applied. Among these modern technologies, machine learning and artificial intelligence, in general, have been at the core of many market prediction models. Specifically, deep learning techniques have been successful in modeling market movements. It has been observed that automatic feature extraction models and time series forecasting techniques have been separately studied. However, a detailed exploration of a stacked framework with multiple different inputs has been lacking. In this whitepaper, we propose a framework based on Convolutional Neural Network (CNN) combined with Long Short-Term Memory (LSTM) to predict the closing price of the Solana cryptocurrency (SOL/USDT). The CNN-LSTM framework extracts features from a rich feature set and applies a time series model with a 30-day trading window to predict the movement of the next day. The model can capture information based on these features to predict the target variable, which is the closing price, with an average absolute percentage error of 2.24% over three years of data. The trading signals based on the proposed deep learning model in this report demonstrate significant improvement in profitability compared to the traditional buy-and-hold strategy.